tirsdag den 6. oktober 2009
Security& privacy in biometrics – how do we ensure proportionality ?
A basic principle in the current European Data Protection Act is to ensure proportionality between the level and amount of personal identifiable data, that you have to reveal to identify yourself has to be proportional to the risk and danger incurred if the identity is faked or stolen.
The recent years have seen a growth in tools for identification, mainly in the biometric area, that has led to the risk of 'overreacting' using easy biometrics where lesser level of authentication could have been used. One of the latest strange cases from Denmark is a night club, that has been allowed by the data protection agency to take customers fingerprints at the entrance as a means to secure against violent behavior. Horror examples of major collection of biometric data is of course U K's collection of DNA profiles for children, a practice that was started 5 or 6 years ago.
The risks involved are related to the kind of threat you are trying to prevent: Do we need the security tool to reveal the identity and all related information? This may be the case if we have a strong suspicion that a person is directly related in crime or an act of terror. Or do we only need to know if a person is 18 years old so it is legal to sell alcohol to him/her? Similarly, within the health area a nurse and a doctor do not need to have full access to a patients medical record if he has lost his consciousness and need a blood transfusion, only the key information of blood type and current medication.
So the use of biometrics in itself is one dimension of the game - and the other dimension is what the biometric identification gives access to reveal of PII – Personally Identifiable Information - at the same time or as a consequence of using the biometrics.
The first question of proportionality is then solely related to the 'strength' of the biometric method used. A weak solution is a quick, convenient solution which is non-intrusive, non-incriminating and non-discriminating in regard to civil rights and color of skin, sex, race and religion. For this purpose simple biometrics like a signature (Analog or digitized) may be better than a fingerprint ( traditional, optical electronic scanning using a template to generate a simple bit stream) - because fingerprints may be seen as incriminating, offensive, police-like. while a face recognition reveals race, color of skin and maybe sex, and thus does not meet the other criteria.
Signatures may be faked, fingerprints (simple fingerprints) can be stolen – in bizarre cases it has been seen that criminals have cut off fingers of owners of Mercedes 300S cars to break the fingerprint starting mechanism. (This risk is probably less in Northern Europe, though.) Or it may be difficult to read the results properly.
When stronger proof is needed, it is acceptable to rely on methods with higher reliability – like the thermal scanning of fingerprints, that measures the distance from the underlying blood, revealing riffs and valleys, again to be transformed by fast fourier transformation to a template consisting of 0's and 1's. This prevents the use of faked fingerprints copied on a strip of tape – and even the rough case of cutting off Mercedes' owner's finger –( presumably the blood has stopped circulating – so no heat difference). Also Iris recognition has been suggested, whereas 3D face recognition at this point still has a higher rate of errors. It has been suggested to use at least 2 types of biometry, like the US border control where you combine fingerprints with face recognition.
In any case the reliability of the identification methodology applied in every case has to discussed and explained before any solution is deployed. (See article about reliability)
It may be OK under well-defined circumstances to use higher level of trusted biometrics, even if they are not 100% proof. The second dimension of the question than is what other PII is stored with the template or the face geometry is stored and how these data are protected. This is a question of data stewardship and again should be in proportion to the use of the data. Taking the example from the Danish night club that has been granted permission to store peoples' fingerprints, these should definitely not be store with any other information than the purpose: Is this guy know to have a tendency to quarrel – NOT his name, address etc. Even if this is kept using cryptography, it is not in proportion to the use of the biometric data.
Other types of biometrics are recognition of moving patterns, voice recognition, pattern of the veins, retina scan – and of course DNA. Whereas the failure rate (both positive and negative) of the first 2 of these types are still relatively high, the 3 other may reveal unwarranted additional details of the health situation of the individual, hence these items should only be used for forensic purposes and not just collected arbitrarily or even – as in the UK DNA case – systematically.
An important aspect of using Biometrics is also how it will be possible to revoke or change the biometrics as the person changes. Whereas fingerprints remain stable for a longer period in life, face geometry changes a lot from childhood to old age, so does walking patterns, voice. And people do have cosmetic operations in their faces, accidents may change the looks and behavior so any system based on biometrics should have a way to allow for changes of this kind and it should be possible to revoke biometrics.
But as the technology improves and computing power is increasing, one solution which could use biometrics and at the same time prevent the data from occurring in the open space or being communicated could be to have an ID card with a number of different domains, each holding the relevant information linked to the person: one domain simply stating the age, another for the bank including bank account numbers, one for driving license use, one for medical/health care use, one for insurance use, one for credit cards, one for public identification purposes.
If this identity card can be activated by a fingerprint reader plus a pin code, the citizen could then select exactly how much PII he wants to reveal in the situation. This is in line with the PrimeLife recommendations from IBM Zürich Lab, that has just got the German award for forward think identity management solution. This type of solution has the advantage that the user is in full control and that no central database is required for the biometric data.
In a few days I will discuss the use of video surveillance, what we know about it as a crime prevention tool and what may be a more intelligent way of using it.